Nonlinear Power System Load Identification Using Local Model Networks

This paper proposes a local model network (LMN) for measurement-based modeling of the nonlinear aggregate power system loads. The proposed LMN approach requires no pre-defined standard load model and uses measurement data to identify load dynamics. Furthermore, due to the interesting characteristics of the proposed approach, the LMN is able to have separate and independent linear and nonlinear inputs, determined by the use of prior knowledge. Trained by the newly developed hierarchical binary tree (HBT) learning algorithm, the proposed LMN attains maximum generalizability with the best linear or nonlinear structure. The previous values of the power system voltage and active and reactive powers are considered as the inputs of the LMN. The proposed approach is applied to the artificially generated data and IEEE 39-bus test system. Work on the field measurement real data is also provided to verify the method. The results of modeling for artificial data, the test system and real data confirm the ability of the proposed approach in capturing the dynamics of the power system loads.